pitch perception in chinchillas ( chinchilla laniger ......chinchillas ( chinchilla laniger ) were...

Pitch Perception in Chinchillas (Chinchilla laniger): StimulusGeneralization Using Rippled Noise

William P. ShofnerIndiana University

William A. Yost and William M. WhitmerLoyola University Chicago

Rippled noises evoke the perception of pitch in human listeners. Infinitely iterated rippled noise (IIRN)is generated when wideband noise (WBN) is delayed, attenuated, and added to the original WBN througheither a positive (�) or a negative (–) feedback loop. The pitch of IIRN[�] is matched to the reciprocalof the delay, whereas the pitch of IIRN[�] for the same delay is an octave lower. Chinchillas (Chinchillalaniger) were trained to discriminate IIRN[�] with a 4-ms delay from IIRN[�] with a 2-ms delay andthen tested in a stimulus generalization paradigm with IIRN[�] at delays between 2 and 4 ms. Systematicgradients in behavioral response occurred along the dimension of delay, suggesting that a perceptualdimension corresponding to pitch exists for IIRN[�]. Behavioral responses to IIRN[�] test stimuli weremore variable among chinchillas, suggesting that IIRN[�] did not evoke similar pitches relative toIIRN[�]. Systematic gradients in behavioral response were observed when IIRN[�] test stimuli werepresented in the context of other IIRN[�] stimuli. Thus, other perceptual cues such as timbre maydominate the pitch cues when IIRN[�] test stimuli are presented in the context of IIRN[�] stimuli.

Keywords: chinchilla, pitch perception, rippled noise, stimulus generalization

A perception of pitch is evoked by many different complexsounds, including speech and music, and different complex soundscan produce the same pitch (Fastl & Stoll, 1979). There exist manydifferent adjectives to describe the variety of pitch perceptionsgenerated by the different types of complex sounds: spectral pitchis generated by simple tones; nonspectral pitch is generated bysinusoidal amplitude modulated noise; periodicity pitch is gener-ated by harmonic complex tones composed of the low-frequency,resolved components; residue pitch is generated by harmoniccomplex tones composed of the high-frequency, unresolved com-ponents; repetition pitch is generated by rippled noise. Ripplednoises are pseudoperiodic sounds; that is, the waveforms possesstemporal regularities, but these regularities are not repeated in aperiodic manner. Consequently, rippled noises have become animportant class of stimuli for studying the perception of pitch andhave proved useful for testing models of pitch perception (Cohen,Grossberg, & Wyse, 1995; Meddis & Hewitt, 1991; Shamma &Klein, 2000; Yost, Patterson, & Sheft, 1996).

Rippled noises are generated when a time-delayed, attenuatedversion of wideband noise (WBN) is added to or subtracted fromthe original WBN. Consequently, the flat-spectrum input is con-

verted into an output stimulus having spectral ripples at frequen-cies related to the delay. A variety of different delay-and-addnetworks can be used to generate rippled noise (see Yost, 1996).Figure 1 illustrates the network used for generating rippled noisesfor the present study, which uses a positive (�) or negative (–)feedback loop. Rippled noises generated from this type of networkare known as infinitely iterated rippled noises (IIRNs) (Shofner &Yost, 1995), comb-filtered noises (Raatgever & Bilsen, 1992), orpeaked ripple noises (Fastl, 1988). For convenience, the parame-ters of IIRN are described using the following notation: IIRN[�/�, T, dBatten], where �/– indicates whether the feedback loop ispositive or negative, T is the delay in milliseconds, and dBatten isthe attenuation of the delayed noise relative to the original, unde-layed version of the noise. Thus, the example spectrum shown inFigure 1 for IIRN[�, 4 ms, �1 dB] is an IIRN generated througha positive feedback loop with a delay of 4 ms and a delayed noiseattenuation of �1 dB.

The pitch perception evoked by the rippled noise is related to thedelay. In general, the pitch of IIRN[�] is at a frequency corre-sponding to 1/T, whereas the pitch of IIRN[�] is at 1/(2T) (Raat-gever & Bilsen, 1992; Yost, 1996); that is, the pitch of IIRN[�] isone octave lower than IIRN[�] of the same delay. The saliency orstrength of the pitch perception is related to the amount of atten-uation in the circuit; as more attenuation is added to the circuit, theresulting rippled noise evokes a stronger noise percept and aweaker pitch percept (e.g., Shofner & Selas, 2002). In the presentstudy, the delay was varied, but the delayed-noise attenuationremained fixed. Thus, for the rippled noises in the present study,the perceptual dimension corresponding to pitch varied, but theperceptual dimension for pitch strength remained relatively con-stant.

Also shown in Figure 1 is an example autocorrelation functionfor the IIRN generated in the positive feedback loop. Positivecorrelations (i.e., peaks) occur in the autocorrelation function at

William P. Shofner, Department of Speech and Hearing Sciences, In-diana University; William A. Yost and William M. Whitmer, ParmlyHearing Institute, Loyola University Chicago.

William M. Whitmer is now at GN ReSound, Glenview, IL.This research was supported by National Institute on Deafness and

Other Communication Disorders Grant R01 DC005596 to William P.Shofner. We thank Noah Jurcin and Angel Losada for technical assistance.

Correspondence concerning this article should be addressed to WilliamP. Shofner, Department of Speech and Hearing Sciences, Indiana Univer-sity, 200 South Jordan Avenue, Bloomington, IN 47405. E-mail:[email protected]

Journal of Comparative Psychology Copyright 2007 by the American Psychological Association2007, Vol. 121, No. 4, 428–439 0735-7036/07/$12.00 DOI: 10.1037/0735-7036.121.4.428

428

integer multiples of the delay, and the peaks in the autocorrelationfunction are an indication of the amount of temporal regularity inthe waveform. The first peak in the autocorrelation function isreferred to as ACpeak1. The time lag where ACpeak1 is located isrelated to the perceived repetition pitch (see Yost, 1996), and inthis example, the time lag of ACpeak1 corresponds to the delay, T.Note that for WBN, the autocorrelation function is flat, indicatingthat no temporal regularities exist in the waveform; consequently,WBN does not evoke a perception of pitch. Figure 2 shows thecomparison of the temporal and spectral characteristics of IIRNsgenerated using positive and negative feedback loops. The top

left-hand panel shows the autocorrelation function for IIRN gen-erated in a positive feedback loop with a delay of 4 ms. Note thatthere are positive correlations (i.e., peaks) at integer multiples ofthe 4-ms delay and that ACpeak1 occurs at 4 ms. The magnitudeor height of ACpeak1 is 0.825 in this example. In the spectrum, thefirst peak occurs at 250 Hz, and all successive peaks are separatedby 250 Hz (see the top right-hand panel of Figure 2). In humanlisteners, this rippled noise evokes a salient pitch of 250 Hz. If anegative feedback loop is used to generate the IIRN with the same4-ms delay, then the autocorrelation function shows alternatingnegative and positive correlations (i.e., nulls and peaks) with the

-120

-100

-80

-60

-40

-20

0

Pow

er (

dB r

e: m

ax)

0 250 500 750 1000 1250 1500Frequency (Hz)

-120

-100

-80

-60

-40

-20

0

Pow

er (

dB r

e: m

ax)

0 250 500 750 1000 1250 1500Frequency (Hz)

WBNIIRN [+/-, T, dB]+/-

Attenuation (dB) Time Delay (T)

IIRN [+, 4 ms, -1 dB]WBN

-1

-0.5

0

0.5

1

Nor

m. A

utoc

orre

latio

n

0 5 10 15 20 25Time (ms)

-1

-0.5

0

0.5

1

Nor

m. A

utoc

orre

latio

n

0 5 10 15 20 25Time (ms)

ACpeak1

Figure 1. Feedback circuit used for generating infinitely iterated rippled noise (IIRN). Example spectra(middle) and autocorrelation functions (bottom) of a wideband noise (WBN) input and IIRN[�, 4 ms, �1 dB]output are shown. The IIRN illustrated was generated using a delay of 4 ms and an attenuation of �1 dB in apositive feedback loop. Norm. � normalized; ACpeak1 � first peak in the autocorrelation function.

429PITCH PERCEPTION IN CHINCHILLAS

first null at the 4-ms delay and the first peak at 8 ms (middleleft-hand panel of Figure 2). Thus, when a negative feedback loopis used, the resulting rippled noise has an ACpeak1 occurring at atime lag equal to twice the delay. In the spectrum, the first peak

occurs at 125 Hz, and successive spectral peaks are separated by250 Hz. In human listeners, this rippled noise evokes a salientpitch of 125 Hz, one octave below that for the positive condition.If a negative feedback loop is used and the delay is at 2 ms, then

-120

-100

-80

-60

-40

-20

0

Pow

er (

dB r

e: m

ax)

0 250 500 750 1000 1250 1500Frequency (Hz)

-120

-100

-80

-60

-40

-20

0P

ower

(dB

re:

max

)

0 250 500 750 1000 1250 1500Frequency (Hz)

-120

-100

-80

-60

-40

-20

0

Pow

er (

dB r

e: m

ax)

0 250 500 750 1000 1250 1500Frequency (Hz)

-1

-0.5

0

0.5

1

Nor

m. A

utoc

orre

latio

n

0 5 10 15 20 25Time (ms)

-1

-0.5

0

0.5

1

Nor

m. A

utoc

orre

latio

n

0 5 10 15 20 25Time (ms)

-1

-0.5

0

0.5

1N

orm

. Aut

ocor

rela

tion

0 5 10 15 20 25Time (ms)

4 ms

4 ms

8 ms

250 Hz

125 Hz

250 Hz

250 Hz

250 Hz

500 Hz

IIRN [+, 4 ms, -1 dB]ACpeak1 = T = 4 ms

IIRN [-, 4 ms, -1 dB]ACpeak1 = 2T = 8 ms

IIRN [-, 2 ms, -1 dB]ACpeak1 = 2T = 4 ms

Figure 2. Examples of the temporal and spectral characteristics of infinitely iterated rippled noises (IIRNs). Top:characteristics for IIRN[�, 4 ms, �1 dB]. Middle: characteristics for IIRN[�, 4 ms, �1 dB]. Bottom: characteristicsfor IIRN[�, 2 ms, �1 dB]. The left-hand columns show autocorrelation functions (temporal); arrows indicate the firstpeak in the autocorrelation function (ACpeak1). The right-hand columns show spectra; single arrows indicate the firstspectral peak, which corresponds to 1/ACpeak1; horizontal double-headed arrows indicate the frequency differencebetween spectral peaks, which corresponds to 1/T, where T is the delay. Norm � normalized.

430 SHOFNER, YOST, AND WHITMER

ACpeak1 occurs at a time lag of 4 ms in the autocorrelationfunction, and the first spectral peak is at 250 Hz with successivepeaks separated by 500 Hz (see bottom panels of Figure 2). Inhuman listeners, this rippled noise also evokes a salient pitch of250 Hz. Although the rippled noises illustrated in the top andbottom panels both evoke matched pitches corresponding to 250Hz, they can be discriminated easily from one another, presumablybecause the spectral differences give rise to timbre differences(e.g., Singh & Hirsh, 1992).

Many of the pitch studies in human listeners have used methodssuch as matching and scaling procedures in order to address issuesconcerned with a participant’s perception rather than with a par-ticipant’s sensory acuity as would be determined by measuringdetection or discrimination thresholds. Although detection anddiscrimination thresholds can be measured in animals, it is notfeasible to use matching or scaling procedures in animals. That is,subjective procedures such as pitch matching and magnitude esti-mation do not have a direct counterpart in animal behavioralexperiments in the way that objective discrimination proceduresdo. However, stimulus generalization paradigms can be used toaddress perceptual questions in animals. In stimulus generalizationparadigms, an animal is trained to respond to a specific trainingstimulus, and then responses are measured to probe or test stimulithat vary along one or more physical dimensions or stimulusfeatures (Mallott & Mallott, 1970). Behavioral responses to teststimuli that are equal to responses to the specific training stimulussuggest a perceptual equivalence among the stimuli (Hulse, 1995).That is, perceptual equivalence occurs when the animal perceivesthere to be a common stimulus feature among the training and teststimuli. If a systematic gradient in behavioral response occurs asthe stimulus dimension changes, then the animal orders the stimulialong the stimulus dimension. This ordering of stimuli along aphysical dimension is known as a generalization gradient, and itpresumably reflects the existence of a perceptual dimension cor-responding to the physical dimension of the stimulus (Guttman,1963). Thus, data from stimulus generalization paradigms canindicate what stimulus features control the behavioral response ofthe animal.

A perceptual dimension corresponding to tone frequency hasbeen established in rats (Blackwell & Schlosberg, 1943), severalspecies of birds (Cynx, 1993; Dooling, Brown, Park, Okanoya, &Soli, 1987; Jenkins & Harrison, 1960), and goldfish (Fay, 1992).Moreover, starlings (Cynx & Shapiro, 1986), cats (Heffner &Whitfield, 1976; Whitfield, 1980), and rhesus monkeys (Tomlin-son & Schwarz, 1988) appear to have a percept corresponding tothe missing fundamental frequency of a complex tone. However,the existence of a perceptual dimension corresponding to ripplednoise delay has only been investigated in goldfish. Using a dis-crimination paradigm, Fay, Yost, and Coombs (1983) showed thatthe thresholds for goldfish are close to those of human listeners fordiscriminating rippled noises of different delays. However, whentrained to detect a rippled noise of one delay and tested withrippled noises in a stimulus generalization paradigm, goldfish didnot show a gradient in behavioral responses (Fay, 2005). In thislatter experiment, Fay (2005) concluded that the lack of a gener-alization gradient in goldfish is because any pitch percept evokedby rippled noise is likely to be weaker than the noise perceptevoked by rippled noise. Thus, whether a perceptual dimension

corresponding to rippled noise delay exists in nonhuman verte-brates remains an open question.

Defining the perceptual characteristics of rippled noises in an-imals is an important conceptual link when making comparisonsbetween the perception of repetition pitch in human listeners andthe responses of single neurons to rippled noise obtained fromanimals (e.g., Shofner, 1999; Wiegrebe & Winter, 2001). Previousstudies have shown that chinchillas discriminate IIRN from WBNusing processes similar to those in human listeners (Shofner &Yost, 1995, 1997) and that chinchillas possess a perceptual dimen-sion related to the periodicity strength of the IIRN (Shofner,Whitmer, & Yost, 2005). However, it is unknown whether aperceptual dimension related to rippled noise delay exists in thechinchilla. In the experiments reported in this article, we usedstimulus generalization procedures to study the perceptual at-tributes of rippled noise delay in the chinchilla.

General Method

Subjects

Adult chinchillas (Chinchilla laniger) served as subjects in theseexperiments. All 5 chinchillas used in the present study had ex-tensive experience in the behavioral paradigm and had served assubjects in previous studies measuring “pitch” strength (Shofner &Whitmer, 2006; Shofner et al., 2005). In these pitch-strengthexperiments, chinchillas discriminated IIRN[�] from a wideband,flat-spectrum noise. Chinchillas received food pellet rewards dur-ing behavioral testing, and their diets were supplemented withchinchilla chow to maintain their body weights at around 80% to90% of normal weight. They received a raisin as a treat followingeach daily behavioral test session. Although chinchillas were foodrestricted, they had free access to water. They were housed indi-vidually in rabbit cages in a room in the animal care facility, andall chinchillas appeared in good health during the period whentheir data were being collected. Chinchillas were tested daily in asession that typically lasted 1 hr.

Acoustic Stimuli

Stimuli consisted of IIRNs as generated by the circuit illustratedin Figure 1. The input WBN to the circuit was generated with aModel 132 VCG/Noise generator (Wavetek, San Diego, CA) inwhich the parameters were set to yield a pseudorandom noise thatrepeated itself every 6.55 s and had a bandwidth of 10 kHz. Thisnoise was then divided into two channels and fed into a digitaldelay line (Model PD 860 Precision Delay Line; Eventide, LittleFerry, NJ). IIRN[�] was generated when the delayed version ofthe WBN was added to the original WBN through a positivefeedback loop, whereas IIRN[�] was generated using a negativefeedback loop. The delay line uses a sampling rate of 62.5 kHz.The outputs of the two delay line channels were low-pass filteredat a cutoff frequency of 15 kHz (FT5 module; Tucker-DavisTechnologies [TDT], Alachua, FL) and summed together (TDTSM3 module). In this study, the amount of attenuation of thedelayed noise was fixed at �1 dB for all IIRNs using a program-mable attenuator (TDT PA4 module). For each IIRN generated, 5 sof the waveform was sampled at 50 kHz and stored as a stimulusfile. For each block of 40 trials during behavioral testing (see


below), a random 500-ms sample of each IIRN was extracted fromthe 5-s stimulus files. Each 500-ms IIRN stimulus was shaped with10-ms rise–fall times.

Stimulus presentation and data acquisition were under the con-trol of a Gateway computer system and TDT System II modules.Stimuli were played through a digital-to-analogue converter (TDTDD1 module) at conversion rate of 50 kHz and low-pass filtered at15 kHz. The output of the low-pass filter was amplified, attenuated(TDT PA4 module), and played through a loudspeaker (RadioShack, Fort Worth, TX). The overall sound pressure level (SPL) ofthe IIRN stimuli was determined by placing a condenser micro-phone (Ivie 1133) at the approximate position of a chinchilla’shead and measuring the A-weighted SPL with a sound level meter(IE-30-A Audio Spectrum Analyzer; Ivie Technologies, Lehi, UT).In this study, the SPL was fixed at 73 dB for all stimuli.

Behavioral Procedure

Chinchillas were placed into a cage (40.6 � 30.5 � 25.4 cm);they were not restrained in any way but were free to roam aroundthe cage. The cage was placed in a single-walled sound-attenuatinganimal test chamber (Industrial Acoustics) that was lined withacoustic foam. A pellet dispenser was located at one end of thecage with a reward chute attached to a response lever. The loud-speaker was placed next to the pellet dispenser approximately 30o

to the right of center at an approximate distance of 6 in. (about 15.2cm) in front of the chinchilla. The behavioral procedure was basedon an operant conditioning paradigm and has been used previouslyto study the perception of periodicity strength in chinchillas(Shofner, 2002; Shofner & Whitmer, 2006; Shofner et al., 2005).It was similar to a procedure used by Ohlemiller, Jones, Heidbre-der, Clark, and Miller (1999) to study categorical perception ofconsonant–vowel syllables in chinchillas.

Figure 3 illustrates the behavioral procedure. A standard stim-ulus was presented continually in 500-ms bursts at a rate of onceper second, regardless of whether or not a trial was initiated.Chinchillas were trained to discriminate a signal stimulus from thestandard stimulus. A trial was initiated when the chinchilla presseddown on the response lever. The lever must be depressed for aspecified duration of time that is referred to as the hold time. Afterthe lever was depressed, the standard stimulus was presented forone to eight bursts. The number of additional bursts was deter-mined for each trial from a rectangular probability distribution andresulted in a random hold time of 1.15 to 8.15 s. If the chinchillareleased the lever before the hold time expired, then the countdownof the hold time was halted; that hold time began again with thenext lever press. If the chinchilla depressed the lever for theduration of the hold time, then one of four stimuli were presentedfor that trial (see Figure 3). The response window was coincidentwith the duration of the trial, which consisted of two 500-ms burstsof a selected stimulus. The response window, however, actuallybegan 150 ms after the onset of the first burst and lasted until theonset of the next burst of the continual standard stimulus. A releaseof the lever during the response window was considered to be apositive response, whereas continuing to depress the lever for theduration of the response window was considered to be a negativeresponse.

A signal trial consisted of two bursts of the signal stimulus. Ifthe chinchilla released the lever during the response window of a

signal trial, then this positive response was treated as a hit. Anegative response during a signal trial was then treated as a miss.A blank trial consisted of two additional bursts of the standardstimulus. If the chinchilla released the lever during the responsewindow of a blank trial, then this positive response was treated asa false alarm. A negative response during a blank trial was treatedas a correct rejection. Hits and correct rejections were correctresponses and were rewarded with food pellets, whereas missesand false alarms were incorrect responses and, as such, were notrewarded with food pellets. That is, positive and negative re-sponses during signal or blank trials were treated as objectiveresponses. A test trial consisted of two bursts of either of twodifferent test or probe stimuli. Chinchillas did not receive foodpellet rewards for responses to test stimuli, regardless of whetherthe behavioral response was positive or negative. Behavioral re-sponses to the test stimuli were considered to be neither correct norincorrect but rather to be subjective responses.

Chinchillas were trained and tested in blocks consisting of 40trials. During periods of training, no test stimuli were presented,and a block of 40 trials consisted of 32 signal trials and 8 standardtrials. During testing sessions, test stimuli were presented infre-quently in the block of trials such that in each block, 60% of thetrials were signal trials (24 of 40 trials), 20% were blank trials (8of 40 trials), 10% were Test Stimulus 1 trials (4 of 40 trials), and10% were Test Stimulus 2 trials (4 of 40 trials). Because chinchil-las only received food pellets for correct responses during signaland blank trials, chinchillas could potentially be rewarded for 80%

Lever Press Random Hold Time(1-8 sec)

Signal Trial(60%)

Blank Trial(20%)

Test #1 Trial(10%)

Test #2 Trial(10%) Press lever for duration

of response window

Release lever duringresponse window

Trial Begins

Press lever for durationof response window


Press lever for durationof response window(Correct Rejection)


(False Alarm)

Press lever for durationof response window

(Miss)


(Hit)

Food Reward

No Reward

No Reward

No Reward

No Reward

No Reward

No Reward

Food Reward

Figure 3. Responses and reinforcement for the stimulus generalizationparadigm. Underlined and italic fonts indicate the correct responses andrewards for each type of trial. sec � seconds.


of the total trials (i.e., 24 signal � 8 blank trials). Behavioralresponses were collected for a minimum of 50 blocks (totaling2,000 trials), which resulted in a minimum of responses to 200trials for each of the two test stimuli presented. Chinchillas typi-cally completed 2 to 8 blocks per day.

Experiment 1: IIRN[�] Tests in the Context of IIRN[�]Comparisons

Method

The purpose of this first experiment was to determine whetherchinchillas possess a perceptual dimension corresponding to pitchthat was related to the delay of IIRN[�]. In this experiment, thestandard stimulus was IIRN[�, 2 ms, �1 dB] and the signalstimulus was IIRN[�, 4 ms, �1 dB]. In human listeners, thisstandard stimulus has a repetition pitch of 500 Hz, whereas thesignal stimulus has a repetition pitch of 250 Hz. Because thechinchillas had previous experience in the behavioral paradigm, itwas only necessary to train the chinchillas to discriminate thesignal stimulus from the standard stimulus. Prior to testing, thechinchillas were trained daily for a period of 1 week. Chinchillaseasily obtained a 90% hit rate within this training period. The teststimuli used in this experiment consisted of IIRN[�] havingdelays (Ts) of 2.22, 2.50, 2.86, and 3.33 ms. Note that forIIRN[�], ACpeak1 is equal to the delay of the rippled noise andthe corresponding repetition pitches will occur at 1/ACpeak1. Forthese test stimuli, the reciprocals of ACpeak1 correspond to 450,400, 350, and 300 Hz, respectively. A systematic gradient inbehavioral responses along the dimension of ACpeak1 (i.e.,T)would be consistent with the existence of a perceptual dimension.This perceptual dimension corresponds to repetition pitch in hu-man listeners.

Results

Figure 4 shows the behavioral responses obtained from 5 chin-chillas when trained to discriminate IIRN[�, 4 ms, �1 dB] fromIIRN[�, 2 ms, �1 dB] and tested with IIRNs having Ts between

2 and 4 ms. The x-axis shows the time lag of ACpeak1 inmilliseconds, which again for IIRN[�] corresponds to T. Behav-ioral response is shown as the percentage of positive responses,which is the percentage of trials on which the chinchilla releasedthe lever during the response window for each particular stimulus.Figure 4 shows that when IIRN[�, 4 ms, �1 dB] was presentedduring the response window (i.e., as ACpeak1 � 4 ms), thepercentage of positive responses was large. That is, there was alarge number of lever releases when the signal stimulus, IIRN[�,4 ms, �1 dB], was presented. In contrast, when IIRN[�, 2 ms, �1dB] was presented during the response window (i.e., ACpeak1 �2 ms), then the percentage of positive responses was small. That is,there were few lever releases when the standard stimulus, IIRN[�,2 ms, �1 dB], was presented. Moreover, when the test stimuliwere presented and there was an increase in the time lag ofACpeak1 from 2 to 4 ms, there was a corresponding systematicincrease in behavioral response. That is, there was a systematicincrease in the number of lever releases during the responsewindow as ACpeak1 of the IIRN stimulus increased from 2 to 4ms. Figure 5A shows the average percentage of positive responsesobtained from the 5 chinchillas as a function of the repetition pitch,1/ACpeak1. The thin line is the best fitting regression line throughthe average data. This figure shows that over this range of delays,which corresponds to an octave change in 1/ACpeak1 from 500 to250 Hz, the average behavioral response can be predicted by alinear function on a semilog scale.


Method

The purpose of this experiment was to determine whetherIIRN[�] evokes a perception corresponding to pitch that is anoctave lower than the corresponding IIRN[�]. In this experiment,the standard stimulus was again IIRN[�, 2 ms, �1 dB] and thesignal stimulus was IIRN[�, 4 ms, �1 dB]. Because the chinchil-las were already trained to make this discrimination for Experi-ment 1, no additional training was necessary. The test stimuli usedin this experiment consisted of IIRN[�] having delays (Ts) of 1.0,1.25, 1.67, and 2.0 ms. Note that for these IIRN[�], ACpeak1 nowoccurs at twice the delay (i.e., at 2T) of the rippled noise. Thecorresponding repetition pitches of 1/ACpeak1 are now at 1/(2T).For example, ACpeak1 for the IIRN[�, 2 ms, �1 dB] test stimuluswill occur at the same time lag as ACpeak1 for the IIRN[�, 4 ms,�1 dB] signal stimulus (compare the top and bottom panels inFigure 2). Thus, for these IIRN[�] test stimuli, the reciprocals ofACpeak1 also correspond to 500, 450, 400, 350, 300, and 250 Hzas in Experiment 1 using only IIRN[�]. A systematic gradient inbehavioral responses along the dimension of ACpeak1 (i.e., 2T)would be consistent with the existence of a perceptual dimensioncorresponding to repetition pitch.

Results

Figure 6 shows the behavioral responses to the IIRN[�] teststimuli in the context of using IIRN[�] as the standard and signalstimuli. As a point of reference, each individual generalizationgradient obtained using IIRN[�] stimuli in Experiment 1 is shown

C16

C29

C40

C41

C43

Mean

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

Figure 4. Percentage of positive responses as a function of the first peakin the autocorrelation function (ACpeak1) for infinitely iterated ripplednoise (IIRN)[�] test stimuli obtained from 5 chinchillas (Cs). The standardstimulus was IIRN[�, 2 ms, �1 dB], and the signal stimulus was IIRN[�,4 ms, �1 dB]. Open symbols with solid lines show gradients fromindividual chinchillas; solid circles with dotted line show the averagepercentage of positive responses across chinchillas.


by the open circles and dotted line in Figure 6. The solid circles inFigure 6 show the responses to the IIRN[�, 2 ms, �1 dB] standardand the IIRN[�, 4 ms, �1 dB] signal stimuli for Experiment 2;these responses are similar to those previously obtained in Exper-iment 1. The gray squares show the behavioral responses to theIIRN[�] test stimuli. Note that for the IIRN[�] test stimuli thedelay and ACpeak1 are not the same. For example, the graysquares in Figure 6 at an ACpeak1 value of 4 ms do not correspondto a signal stimulus with a delay of 4 ms but rather correspond tothe IIRN[�] test stimulus having a delay of 2 ms. All of the graysquares in Figure 6 indicate responses to IIRN[�] test stimuli.

Behavioral responses to IIRN[�] that are equivalent to IIRN[�]having the same ACpeak1 would suggest a similarity in the evokedpitches. That is, if the behavioral responses to IIRN[�] test stimulifall along the IIRN[�] generalization gradient (the dotted line inFigure 6), then it would suggest a similarity in evoked pitchesbetween IIRN[�] and IIRN[�] stimuli having the same ACpeak1.For the most part, the responses to IIRN[�] test stimuli did not fallalong the generalization gradients for IIRN[�] stimuli. Most of the

responses to IIRN[�] stimuli were lower than those of the corre-sponding IIRN[�] stimuli, suggesting that the IIRN[�] were notperceived to be similar to the corresponding IIRN[�]. The largestbehavioral responses were obtained for IIRN[�] having an AC-peak1 at 4 ms, but note that these responses were typically wellbelow the responses obtained to IIRN[�] at the same ACpeak1(see the double-headed arrows in Figure 6).


Method

The lack of a systematic gradient in behavioral responses forIIRN[�] shown in Figure 6 could reflect an absence of a pitchpercept evoked by IIRN[�] stimuli (see the Discussion section).That is, perhaps in the chinchilla, IIRN[�] stimuli do not evoke aperceptual dimension corresponding to repetition pitch the waythat IIRN[�] stimuli do (i.e., Figure 4). Alternatively, it is possiblethat IIRN[�] stimuli do evoke a perceptual dimension correspond-ing to pitch, but the spectral differences between IIRN[�] andIIRN[�] stimuli evoke a perceptual dimension corresponding totimbre that dominates any pitch percept. In order to test this, weobtained generalization gradients in which IIRN[�, 1 ms, �1 dB]was used as the standard stimulus and IIRN[�, 2 ms, �1 dB] wasused as the signal stimulus. Chinchillas were then tested withIIRN[�] with Ts of 1.11, 1.25, 1.43, and 1.67 ms. For Chinchillas16 and 29, no training was necessary as the chinchillas achieved hitrates of 90% on the 1st day; Chinchillas 40 and 41 were given 3weeks and 1 week of training, respectively, before testing began.

Results

Unlike the responses observed in Figure 6 in which IIRN[�]stimuli were tested in the context of IIRN[�] standard and signalstimuli, Figure 7 shows the responses to IIRN[�] test stimuliobtained in the context of IIRN[�] standard and signal stimuli.Three of the 4 chinchillas tested (Chinchillas 16, 40, and 41)appeared to show a systematic increase in percentage of positiveresponses as ACpeak1 of the IIRN[�] stimuli increased from 2 to4 ms. Comparison of the behavioral responses to IIRN[�] withthose to IIRN[�] (solid circles vs. open circles in Figure 7)indicates that the generalization gradients are similar for the twosets of stimuli. One chinchilla (Chinchilla 29) gave a generaliza-tion gradient for IIRN[�] that was not as systematic and deviatedmore from the IIRN[�] gradient.

A comparison of the behavioral responses as a function ofrepetition pitch is shown in Figure 5 for IIRNs generated using thepositive and negative feedback loops. Repetition pitch is defined as1/ACpeak1 in hertz. These functions show the mean positiveresponses and the 95% confidence intervals. As the repetition pitchof the IIRN increased from 250 Hz, there was a systematic de-crease in the percentage of positive responses for both IIRN[�](see Figure 5A) and IIRN[�] (see Figure 5B). The data can be wellfit with linear functions. The slope of the best fitting IIRN[�]regression line is �297, and the y-intercept is 805 (Figure 5A); theslope of the best fitting IIRN[�] regression line is �315, and they-intercept is 845 (Figure 5B). The r2 values for the IIRN[�] andIIRN[�] regression lines are .985 and .970, respectively.

-20

0

20

40

60

80

100

120

0001001

1/ACpeak1 (Hz)

% P

os

itiv

e R

es

po

ns

es

-20

0

20

40

60

80

100

120

0001001

1/ACpeak1 (Hz)

% P

os

itiv

e R

es

po

ns

es

IIRN [+, T, -1 dB]

IIRN [-, T, -1 dB]

A

B

Figure 5. Percentage of positive responses as a function of repetitionpitch, 1/ACpeak1, where ACpeak1 is the first peak in the autocorrelationfunction. Error bars indicate the 95% confidence intervals based onTukey’s standard error. A: Mean responses of 5 chinchillas trained todiscriminate infinitely iterated rippled noise (IIRN)[�, 4 ms, �1 dB] fromIIRN[�, 2 ms, �1 dB] and tested with IIRN[�] stimuli with delaysbetween 2 and 4 ms. B: Mean responses of four chinchillas trained todiscriminate IIRN[�, 2 ms, �1 dB] from IIRN[�, 1 ms, �1 dB] and testedwith IIRN[�] stimuli with delays between 1 and 2 ms. The solid lines showthe best fitting regression lines through the means.


Discussion

Behavioral responses were measured from chinchillas using astimulus generalization paradigm to IIRNs that varied in delay.Stimulus generalization paradigms are designed to address ques-tions more concerned with measuring perception in animals ratherthan with measuring sensory acuity as in detection and discrimi-nation paradigms. In a stimulus generalization paradigm, animalsare trained to discriminate a specific training or signal stimulusfrom a comparison or standard stimulus. Behavioral responses arethen measured for probe or test stimuli that vary systematicallyalong one or more stimulus dimensions (Mallott & Mallott, 1970).

Systematic changes in behavioral responses along a stimulus di-mension are referred to as stimulus generalization gradients andare interpreted to be a reflection of the psychological or perceptualdimension of the stimulus (Guttman, 1963). Behavioral responsesto test stimuli that are similar in magnitude to responses to thesignal stimulus are interpreted to indicate a perceptual equivalenceamong the stimuli (Hulse, 1995); that is, from a functional view-point, these stimuli contain a feature that makes the animal per-ceive the stimuli to be equivalent or similar.

In the present study, the stimulus dimension in question is thedelay of the rippled noise, or more specifically ACpeak1. The

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C16

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C29

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C40

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C43

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C41

IIRN [+, 2 ms] = standardIIRN [+, 4 ms] = signalIIRN [+, T] = test

IIRN [+, 2 ms] = standardIIRN [+, 4 ms] = signalIIRN [ -, T] = test

53.4%

25.9%5.4%

30.5%14.9%

Figure 6. Percentage of positive responses as a function of the first peak in the autocorrelation function(ACpeak1) for infinitely iterated rippled noise (IIRN)[�] test stimuli obtained from 5 chinchillas (Cs). Thestandard stimulus was IIRN[�, 2 ms, �1 dB], and the signal stimulus was IIRN[�, 4 ms, �1 dB]. Open circleswith dotted line are generalization gradients obtained from Experiment 1 and shown in Figure 4. Solid circlesshow the percentage of positive responses to the IIRN[�, 2 ms, �1 dB] standard and the IIRN[�, 4 ms, �1 dB]signal when tested with IIRN[�] stimuli. Percentage of positive responses to IIRN[�] are shown in thegray-shaded squares. The vertical double-headed arrows show the difference between the responses to theIIRN[�, 4 ms, �1 dB] signal stimulus and the IIRN[�, 2 ms, �1 dB] test stimulus. Both of these haveACpeak1s of 4 ms. The percentages indicate the amount that the responses to IIRN[�, 2 ms, �1 dB] decreasedrelative to that for the IIRN[�, 4 ms, �1 dB].


repetition pitch is directly related to the time lag of ACpeak1 inhuman listeners (Yost, 1996). Chinchillas were trained to discrim-inate an IIRN[�] signal with an ACpeak1 at 4 ms from anIIRN[�] standard with an ACpeak1 at 2 ms. In human listeners,this represents an octave change in pitch from 500 Hz (standard) to250 Hz (signal). Chinchillas show monotonic changes in behav-ioral response as ACpeak1 of IIRN[�] changes from 2 to 4 ms(Experiment 1). That is, chinchillas systematically order theseIIRN[�] stimuli along a physical dimension of ACpeak1. Figure5A shows that the behavioral response of the chinchilla as afunction of the reciprocal of ACpeak1 (i.e., frequency in hertz) isrepresented well by a straight line when plotted in linear-logcoordinates. The generalization gradient in the chinchilla is con-sistent with the existence of a perceptual dimension that corre-

sponds to rippled noise pitch (i.e., repetition pitch). That is, therepetition pitch scale over the octave range of 250 to 500 Hz forchinchillas is a linear function on a linear-log coordinate system. Itis interesting to note that the shape of the pure tone pitch scale forhumans and budgerigars (Dooling et al., 1987) and in starlings(Cynx, 1993) over a narrow frequency range on the order of oneoctave is also linear in a linear-log coordinate system. Nonlinearpitch scales are obtained when the frequency range is severaloctaves wide (Blackwell & Schlosberg, 1943; Dooling et al.,1987).

Although rippled noise processing has been examined in birds(Amagai, Dooling, Shamma, Kidd, & Lohr, 1999) and mammals(e.g., Shofner & Yost, 1995; Shofner et al., 2005), these studieshave not been concerned with pitch processing per se but rather

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C29

0

20

40

60

80

100%

Pos

itive

Res

pons

es

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C16

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C40

0

20

40

60

80

100

% P

ositi

ve R

espo

nses

1.5 2.0 2.5 3.0 3.5 4.0 4.5ACpeak1 (ms)

C41

Stimulus Generalization

IIRN [+, 2 ms] = standardIIRN [+, 4 ms] = signalIIRN [+, T] = test

IIRN [ -, 1 ms] = standardIIRN [ -, 2 ms] = signalIIRN [ -, T] = test

Figure 7. Comparison of behavioral responses for infinitely iterated rippled noise (IIRN)[�] and IIRN[�]stimuli obtained from 4 chinchillas (Cs) in the stimulus generalization procedure. Each panel shows generali-zation gradients as the percentage of positive responses as a function of the first peak in the autocorrelationfunction (ACpeak1). Solid circles with solid lines show generalization gradients obtained from when thestandard stimulus was IIRN[�, 1 ms, �1 dB] and the signal stimulus was IIRN[�, 2 ms, �1 dB]; test stimuliwere IIRN[�] having delays (Ts) of 1.11, 1.25, 1.43, and 1.67 ms. Open circles with dotted lines show gradientsobtained from Experiment 1 when the standard stimulus was IIRN[�, 2 ms, �1 dB] and the signal stimulus wasIIRN[�, 4 ms, �1 dB]; test stimuli were IIRN[�] having Ts of 2.22, 2.5, 2.86, and 3.33 ms. Note that C41developed a prolapsed bowel before collection of the data was completed for two of the IIRN[�] test stimuli withTs of 1.11 and 1.43 ms (i.e., ACpeak1s of 2.22 and 2.86 ms, respectively). Consequently, this chinchilla wasremoved from the remainder of the study.


have addressed issues regarding periodicity strength of ripplednoises. That is, the discriminations were between a rippled noiseand a wideband, flat-spectrum noise (i.e., coloration discrimina-tion) rather than between rippled noises of different delays (i.e.,pitch discrimination). There is one other study in which the exis-tence of a perceptual dimension corresponding to rippled noisedelay was addressed. In goldfish, Fay (2005) examined the exis-tence of a pitchlike dimension for rippled noise stimuli usingstimulus generalization. This study showed that goldfish general-ized to all rippled noise stimuli when trained to detect a ripplednoise signal having a specific delay. That is, goldfish responded toall test rippled noises with the same probability as their response tothe signal rippled noise and, consequently, did not produce ageneralization gradient. The lack of a generalization gradient issuggestive that a perceptual dimension corresponding to ripplednoise delay is weak or nonexistent in goldfish (Fay, 2005). Nev-ertheless, goldfish do appear to possess pitchlike perceptual di-mensions corresponding to tone frequency (Fay, 1992) and thefundamental frequency of complex tones (Fay, 2005). The noisepercept of the rippled noise appears to overwhelm any pitchpercept evoked by the rippled noise (Fay, 2005). The stimulusgeneralization gradients along the physical dimension of ACpeak1obtained in the present study are the first reported data indicatingthat a perceptual dimension corresponding to rippled noise delayexists in nonhuman vertebrates.

When chinchillas were trained using IIRN[�] stimuli and testedusing IIRN[�] stimuli (Experiment 2), monotonic changes inbehavioral response were not observed as ACpeak1 changed from2 to 4 ms. That is, chinchillas did not order IIRN[�] test stimulisystematically along the physical dimension of ACpeak1. The lackof a generalization gradient could be interpreted as a lack of apitchlike perception for IIRN[�] corresponding to the reciprocalof ACpeak1. That is, IIRN[�] stimuli may not be perceptuallyequivalent to IIRN[�] stimuli having the same values of ACpeak1and, thus, do not evoke a pitchlike perception in chinchillas. Whatmight control the behavioral response of the chinchilla in thisexperiment if it is not the location of ACpeak1? First, the magni-tude of ACpeak1 could affect the behavioral response, and second,the spectral differences (see Figure 2) could affect the behavioralresponse.

It has previously been shown that the magnitude of ACpeak1 ofrippled noise is related to the perceptual dimension correspondingto pitch strength in chinchillas (Shofner et al., 2005) as well as inhumans (Shofner & Selas, 2002). However, pitch strength seems tobe an unlikely cue for controlling the behavioral responses of thechinchilla in the present study for two reasons. First, the magnitudeof ACpeak1 for the comparison stimuli (i.e., IIRN[�, 4 ms, �1dB] and IIRN[�, 2 ms, �1 dB]) are both around a value of 0.8 (seeFigure 2 for IIRN[�, 4 ms, �1 dB] example). Thus, if thediscrimination between this standard stimulus and this signal stim-ulus was solely based on pitch strength, then these two stimulicould not be discriminated and the behavioral responses would beequal. Second, as can be observed in Figure 6 (solid circles), thebehavioral responses obtained to the standard were low, whereasthe behavioral responses obtained to the signal were high, suggest-ing that these rippled noises are easily discriminated by the chin-chilla. The magnitudes of ACpeak1 for the IIRN[�] test stimuliranged from 0.614 to 0.720 with a median value of 0.631. Thismedian value is close to the magnitude of ACpeak1 for IIRN[�]

stimuli with a delayed noise attenuation of �3 dB (see Figure 2 ofShofner, 2002). When IIRN[�] was used as the signal stimulusand delayed noise attenuation was varied from �1 dB to �3 dB,the behavioral responses decreased from greater than or equal to90% to around 50% to 60% (see Shofner et al., 2005). If thebehavioral responses in the present experiment were being con-trolled solely by pitch strength, then it would be expected thatbehavioral responses would be around 50% for all IIRN[�] teststimuli. Figure 6 clearly shows that this is not the case; behavioralresponses to IIRN[�] test stimuli were not uniformly around 50%.Thus, the behavioral responses to IIRN[�] test stimuli in thecontext of IIRN[�] comparison stimuli do not appear to be con-trolled by the perceptual dimension corresponding to pitchstrength.

A perceptual dimension corresponding to timbre could have aninfluence on the behavioral responses to IIRN[�] test stimuli inthe context of IIRN[�] comparison stimuli. For example, considerIIRN[�, 4 ms, �1 dB] and IIRN[�, 2 ms, �1 dB] as illustratedin the top and bottom panels of Figure 2. Both stimuli sharecommon features, namely, an ACpeak1 corresponding to a timelag of 4 ms and a first spectral peak located at 250 Hz. However,there are considerable spectral differences between the two stimuli.IIRN[�, 4 ms, �1 dB] has spectral peaks at all integer multiplesof 250 Hz, whereas IIRN[�, 2 ms, �1 dB] has spectral peaks onlyat the odd integer multiples of 250 Hz. For complex sounds that areequal in loudness and pitch, differences in timbre can occur whenthere are spectral differences. Singh and Hirsh (1992) showed thatfor human listeners, timbre is influenced primarily by spectrallocus, whereas pitch is influenced primarily by the fundamentalfrequency. Thus, the spectral differences described above forIIRN[�] and IIRN[�] stimuli could evoke a perception corre-sponding to timbre in the chinchilla. In the present study, the lackof a generalization gradient to IIRN[�] stimuli in the context ofIIRN[�] comparison stimuli may reflect a perception correspond-ing to timbre. That is, the behavioral response of the chinchilla toIIRN[�] in the context of IIRN[�] may have been under thecontrol of spectral differences between IIRNs rather than temporalsimilarities between IIRNs.

To gain some insight into whether the behavioral responses ofchinchillas to IIRN[�] in the context of IIRN[�] stimuli werecontrolled by pitchlike cues or timbrelike cues, we tested chinchil-las with IIRN[�] stimuli using IIRN[�] standard and signalstimuli. That is, responses to IIRN[�] test stimuli were nowmeasured in the context of IIRN[�] comparison stimuli. Thus, inthis experiment, pitch differences between the comparison and teststimuli should remain, whereas timbre differences between com-parison and test stimuli should be eliminated or reduced. That is,in Experiment 3 there were systematic differences in the locationof ACpeak1 and all stimuli had spectral peaks only at odd integermultiples of 1/(2T).

When chinchillas were trained using IIRN[�] comparison stim-uli and tested with IIRN[�] stimuli (Experiment 3), monotonicchanges in behavioral response were now observed as ACpeak1changed from 2 to 4 ms. That is, most chinchillas showed mono-tonic changes in behavioral response as ACpeak1 of IIRN[�]changed from 2 to 4 ms, indicating that chinchillas systematicallyorder these IIRN[�] stimuli along a physical dimension ofACpeak1. One chinchilla (Chinchilla 29) did show a generaliza-tion gradient that was not as orderly as that obtained for IIRN[�]


stimuli. For most chinchillas, however, the generalization gradi-ents for IIRN[�] were similar to those for IIRN[�] (see Figure 7).The IIRN[�] generalization gradient in the chinchilla is consistentwith the existence of a perceptual dimension that also correspondsto rippled noise pitch. Figure 5 shows that the behavioral responseof the chinchilla as a function of the reciprocal of ACpeak1 (i.e.,frequency in hertz) is represented well by a straight line whenplotted in linear-log coordinates for IIRN[�] stimuli as well asIIRN[�] stimuli. Comparison of the two regression lines suggeststhat the perceptual dimension corresponding to a repetition pitchscale over the octave range of 250 to 500 Hz for chinchillas issimilar for IIRN[�] and IIRN[�] stimuli.

The systematic ordering of IIRN[�] test stimuli along a phys-ical dimension of ACpeak1 in the context of IIRN[�] comparisonstimuli suggests that IIRN[�] stimuli do indeed evoke a pitchlikeperception in chinchillas. However, when IIRN[�] is placed in thecontext of IIRN[�] stimuli as in Experiment 2, the perceptualdimension that corresponds to repetition pitch appears to be dom-inated by the perceptual dimension corresponding to timbre. Thus,in the context of IIRN[�] comparison stimuli, the behavioralresponse of chinchillas to IIRN[�] is controlled by timbre cuesrather than pitch cues. The dominance of timbre cues over pitchcues in this context has also been observed in other studies. Forexample, when goldfish are trained to respond to a pure tone, theydo not generalize to any IIRN[�] test stimuli, even to those ripplednoises that would evoke the same pitch as the tone in humanlisteners (Fay, 2005). That is, IIRN[�] stimuli are not perceptuallyequivalent to tones. Presumably, in the context of pure tones,IIRN[�] test stimuli evoke more of a noise percept than a pitchpercept in goldfish, and consequently, the behavioral responses ofthe goldfish to IIRN[�] test stimuli are controlled by timbrelikecues rather than pitchlike cues. In human listeners, interactionsbetween pitch and timbre can occur such that the timbre of a soundcan have an influence on the perceived pitch of the sound (e.g.,Singh & Hirsh, 1992; Warrier & Zatorre, 2002).

The results presented in this article indicate that a perceptualdimension corresponding to repetition pitch exists in chinchillas. Achange in rippled noise delay has an effect on the behavioralresponse of the chinchilla, and, more important, the generalizationgradient changes systematically as delay changes. A change in therippled noise delay evokes a change in repetition pitch in humanlisteners. If a generalization gradient reflects the existence of aperceptual dimension corresponding to the physical dimension ofthe stimulus as argued by Guttman (1963), then a change in thedelay of rippled noise also gives rise to a corresponding perceptualdimension in chinchillas.

References

Amagai, S., Dooling, R. J., Shamma, S., Kidd, T. L., & Lohr, B. (1999).Detection of modulation in spectral envelopes and linear-rippled noisesby budgerigars (Melopsittacus undulates). Journal of the AcousticalSociety of America, 105, 2029–2035.

Blackwell, H. R., & Schlosberg, H. (1943). Octave generalization, pitchdiscrimination, and loudness thresholds in the white rat. Journal ofExperimental Psychology, 33, 407–419.

Cohen, M. A., Grossberg, S., & Wyse, L. (1995). A spectral network modelof pitch perception. Journal of the Acoustical Society of America, 98,862–879.

Cynx, J. (1993). Auditory frequency generalization and a failure to find

octave generalization in a songbird, the European starling (Sturnusvulgaris). Journal of Comparative Psychology, 107, 140–146.

Cynx, J., & Shapiro, M. (1986). Perception of missing fundamental by aspecies of songbird (Sturnus vulgaris). Journal of Comparative Psychol-ogy, 100, 356–402.

Dooling, R. J., Brown, S. D., Park, T. J., Okanoya, K., & Soli, S. D. (1987).Perceptual organization of acoustic stimuli by budgerigars (Melopsitta-cus undulates): I. Pure tones. Journal of Comparative Psychology, 101,139–149.

Fastl, H. (1988). Pitch and pitch strength of peaked ripple noise. In H.Duifhuis, J. W. Horst, & H. P. Wit (Eds.), Basic issues in hearing (pp.370–379). San Diego, CA: Academic Press.

Fastl, H., & Stoll, G. (1979). Scaling of pitch strength. Hearing Research,1, 293–301.

Fay, R. R. (1992). Analytic listening by the goldfish. Hearing Research,59, 101–107.

Fay, R. R. (2005). Perception of pitch by goldfish. Hearing Research, 205,7–20.

Fay, R. R., Yost, W. A., & Coombs, S. (1983). Psychophysics andneurophysiology of repetition noise processing in a vertebrate auditorysystem. Hearing Research, 12, 31–55.

Guttman, N. (1963). Laws of behavior and facts of perception. In S. Koch(Ed.), Psychology: A study of science: Vol. 5. The process areas, theperson, and some applied fields: Their place in psychology and science(pp. 114–178). New York: McGraw-Hill.

Heffner, H., & Whitfield, I. C. (1976). Perception of the missing funda-mental by cats. Journal of the Acoustical Society of America, 59,915–919.

Hulse, S. H. (1995). The discrimination-transfer procedure for studyingauditory perception and perceptual invariance in animals. In G. M.Klump, R. J. Dooling, R. R. Fay, & W. C. Stebbins (Eds.), Methods incomparative psychoacoustics (pp. 319 –330). Basel, Switzerland:Birkhauser Verlag.

Jenkins, H. M., & Harrison, R. H. (1960). Effect of discrimination trainingon auditory generalization. Journal of Experimental Psychology, 59,246–253.

Mallott, R. W., & Mallott, M. K. (1970). Peception and stimulus general-ization. In W. C. Stebbins (Ed.), Animal psychophysics: The design andconduct of sensory experiments (pp. 363–400). New York: Appleton-Century-Crofts.

Meddis, R., & Hewitt, M. J. (1991). Virtual pitch and phase sensitivity ofa computer model of the auditory periphery. Journal of the AcousticalSociety of America, 89, 2866–2882.

Ohlemiller, K. K., Jones, L. B., Heidbreder, A. F., Clark, W. W., & Miller,J. D. (1999). Voicing judgements by chinchillas trained with a rewardparadigm. Behavioural Brain Research, 100, 185–195.

Raatgever, J., & Bilsen, F. A. (1992). The pitch of anharmonic combfiltered noise reconsidered. In Y. Cazals, K. Horner, & L. Demany(Eds.), Auditory physiology and perception (pp. 215–222). New York:Pergamon Press.

Shamma, S., & Klein, D. (2000). The case of the missing pitch templates:How harmonic templates emerge in the early auditory system. Journal ofthe Acoustical Society of America, 107, 2631–2644.

Shofner, W. P. (1999). Responses of cochlear nucleus units in the chin-chilla to iterated rippled noises: Analysis of neural autocorrelograms.Journal of Neurophysiology, 81, 2662–2674.

Shofner, W. P. (2002). Perception of the periodicity strength of complexsounds by the chinchilla. Hearing Research, 173, 69–81.

Shofner, W. P., & Selas, G. (2002). Pitch strength and Stevens’s powerlaw. Perception & Psychophysics, 64, 437–450.

Shofner, W. P., & Whitmer, W. M. (2006). Pitch cue learning in chinchil-las: The role of spectral region in the training stimulus. Journal of theAcoustical Society of America, 120, 1706–1712.

Shofner, W. P., Whitmer, W. M., & Yost, W. A. (2005). Listening


experience with iterated rippled noise alters the perception of “pitch”strength of complex sounds in the chinchilla. Journal of the AcousticalSociety of America, 118, 3187–3197.

Shofner, W. P., & Yost, W. A. (1995). Discrimination of rippled-spectrumnoise from flat-spectrum wideband noise by chinchillas. Auditory Neu-roscience, 1, 127–138.

Shofner, W. P., & Yost, W. A. (1997). Discrimination of rippled-spectrumnoise from flat-spectrum wideband noise by chinchillas: Evidence for aspectral dominance region. Hearing Research, 110, 15–24.

Singh, P. G., & Hirsh, I. J. (1992). Influence of spectral locus and F0changes on the pitch and timbre of complex tones. Journal of theAcoustical Society of America, 92, 2650–2661.

Tomlinson, R. W. W., & Schwarz, D. W. F. (1988). Perception of themissing fundamental in nonhuman primates. Journal of the AcousticalSociety of America, 84, 560–565.

Warrier, C. M., & Zatorre, R. J. (2002). Influence of tonal context and

timbral variation on perception of pitch. Perception & Psychophysics,64, 198–207.

Whitfield, I. C. (1980). Auditory cortex and the pitch of complex tones.Journal of the Acoustical Society of America, 67, 644–647.

Wiegrebe, L., & Winter, I. M. (2001). Temporal representation of iteratedrippled noise as a function of delay and sound level in the ventralcochlear nucleus. Journal of Neurophysiology, 85, 1206–1219.

Yost, W. A. (1996). Pitch of iterated rippled noise. Journal of the Acous-tical Society of America, 100, 511–518.

Yost, W. A., Patterson, R. D., & Sheft, S. (1996). A time domain descrip-tion for the pitch strength of iterated rippled noise. Journal of theAcoustical Society of America, 99, 1066–1078.

Received September 22, 2006Revision received May 22, 2007

Accepted May 24, 2007 �


pitch perception in chinchillas ( chinchilla laniger ......chinchillas ( chinchilla laniger ) were...

Documents